Our JointDiff and JointDiff-x models were originally developed for protein sequence–structure co-design of monomer proteins. Both models employ a multimodal diffusion architecture to jointly update sequence and structure. Specifically, JointDiff follows the traditional DDPM framework, predicting noise and iteratively denoising samples, while JointDiff-x directly predicts the ground truth and refines samples progressively.
For this task, we extended these unconditional monomer design models to binder design, enabling conditioning on specific target proteins. The models were further trained on protein complex data derived from ProteinMPNN. For the generated designs, we evaluated their diversity and novelty, and ranked them based on the ipSAE score.
id: vast-jaguar-rose

RBX1
None
37.83
False
10.9 kDa
100
id: gentle-swan-thorn

RBX1
None
63.11
True
11.3 kDa
100
id: rapid-bee-pine

RBX1
None
42.60
False
10.0 kDa
100
id: dark-kiwi-frost

RBX1
None
48.81
False
10.3 kDa
100
id: pale-panda-vine

RBX1
None
67.91
True
11.8 kDa
100
id: solid-bee-wave

RBX1
None
65.12
True
11.6 kDa
100
id: frozen-lion-sand

RBX1
None
32.58
False
10.6 kDa
100
id: hollow-kiwi-oak
No preview available
--
--
--
--
--
100
id: soft-orca-lava
No preview available
--
--
--
--
--
150
id: deep-deer-maple
No preview available
--
--
--
--
--
100
id: crimson-falcon-fern
No preview available
--
--
--
--
--
150
id: young-ibis-oak
No preview available
--
--
--
--
--
100
id: swift-kiwi-plume
No preview available
--
--
--
--
--
150
id: pale-jaguar-opal
No preview available
--
--
--
--
--
150
id: green-crane-frost
No preview available
--
--
--
--
--
100
id: rough-raven-dust
No preview available
--
--
--
--
--
150
id: ivory-mole-dust
No preview available
--
--
--
--
--
200
id: wild-bear-oak
No preview available
--
--
--
--
--
150
id: silver-owl-sand
No preview available
--
--
--
--
--
150
id: golden-cobra-cypress
No preview available
--
--
--
--
--
200
id: rough-hawk-iron
No preview available
--
--
--
--
--
100